Dev gzf exp (#1654)

zhifu gao

2024-04-24 861147c7308b91068ffa02724fdf74ee623a909e

 runtime/python/websocket/funasr_wss_client.py

@@ -3,11 +3,13 @@
import time
import websockets, ssl
import asyncio

# import threading
import argparse
import json
import traceback
from multiprocessing import Process

# from funasr.fileio.datadir_writer import DatadirWriter

import logging
@@ -15,72 +17,34 @@
logging.basicConfig(level=logging.ERROR)

parser = argparse.ArgumentParser()
parser.add_argument("--host",
                    type=str,
                    default="localhost",
                    required=False,
                    help="host ip, localhost, 0.0.0.0")
parser.add_argument("--port",
                    type=int,
                    default=10095,
                    required=False,
                    help="grpc server port")
parser.add_argument("--chunk_size",
                    type=str,
                    default="5, 10, 5",
                    help="chunk")
parser.add_argument("--encoder_chunk_look_back",
                    type=int,
                    default=4,
                    help="chunk")
parser.add_argument("--decoder_chunk_look_back",
                    type=int,
                    default=0,
                    help="chunk")
parser.add_argument("--chunk_interval",
                    type=int,
                    default=10,
                    help="chunk")
parser.add_argument("--hotword",
parser.add_argument(
    "--host", type=str, default="localhost", required=False, help="host ip, localhost, 0.0.0.0"
)
parser.add_argument("--port", type=int, default=10095, required=False, help="grpc server port")
parser.add_argument("--chunk_size", type=str, default="5, 10, 5", help="chunk")
parser.add_argument("--encoder_chunk_look_back", type=int, default=4, help="chunk")
parser.add_argument("--decoder_chunk_look_back", type=int, default=0, help="chunk")
parser.add_argument("--chunk_interval", type=int, default=10, help="chunk")
parser.add_argument(
    "--hotword",
                    type=str,
                    default="",
                    help="hotword file path, one hotword perline (e.g.:阿里巴巴 20)")
parser.add_argument("--audio_in",
                    type=str,
                    default=None,
                    help="audio_in")
parser.add_argument("--audio_fs",
                    type=int,
                    default=16000,
                    help="audio_fs")
parser.add_argument("--send_without_sleep",
    help="hotword file path, one hotword perline (e.g.:阿里巴巴 20)",
)
parser.add_argument("--audio_in", type=str, default=None, help="audio_in")
parser.add_argument("--audio_fs", type=int, default=16000, help="audio_fs")
parser.add_argument(
    "--send_without_sleep",
                    action="store_true",
                    default=True,
                    help="if audio_in is set, send_without_sleep")
parser.add_argument("--thread_num",
                    type=int,
                    default=1,
                    help="thread_num")
parser.add_argument("--words_max_print",
                    type=int,
                    default=10000,
                    help="chunk")
parser.add_argument("--output_dir",
                    type=str,
                    default=None,
                    help="output_dir")
parser.add_argument("--ssl",
                    type=int,
                    default=1,
                    help="1 for ssl connect, 0 for no ssl")
parser.add_argument("--use_itn",
                    type=int,
                    default=1,
                    help="1 for using itn, 0 for not itn")
parser.add_argument("--mode",
                    type=str,
                    default="2pass",
                    help="offline, online, 2pass")
    help="if audio_in is set, send_without_sleep",
)
parser.add_argument("--thread_num", type=int, default=1, help="thread_num")
parser.add_argument("--words_max_print", type=int, default=10000, help="chunk")
parser.add_argument("--output_dir", type=str, default=None, help="output_dir")
parser.add_argument("--ssl", type=int, default=1, help="1 for ssl connect, 0 for no ssl")
parser.add_argument("--use_itn", type=int, default=1, help="1 for using itn, 0 for not itn")
parser.add_argument("--mode", type=str, default="2pass", help="offline, online, 2pass")

args = parser.parse_args()
args.chunk_size = [int(x) for x in args.chunk_size.split(",")]
@@ -102,6 +66,7 @@
async def record_microphone():
    is_finished = False
    import pyaudio

    # print("2")
    global voices
    FORMAT = pyaudio.paInt16
@@ -112,11 +77,9 @@

    p = pyaudio.PyAudio()

    stream = p.open(format=FORMAT,
                    channels=CHANNELS,
                    rate=RATE,
                    input=True,
                    frames_per_buffer=CHUNK)
    stream = p.open(
        format=FORMAT, channels=CHANNELS, rate=RATE, input=True, frames_per_buffer=CHUNK
    )
    # hotwords
    fst_dict = {}
    hotword_msg = ""
@@ -141,7 +104,9 @@
    if args.use_itn == 0:
        use_itn=False
    
    message = json.dumps({"mode": args.mode,
    message = json.dumps(
        {
            "mode": args.mode,
                          "chunk_size": args.chunk_size,
                          "chunk_interval": args.chunk_interval,
                          "encoder_chunk_look_back": args.encoder_chunk_look_back,
@@ -150,7 +115,8 @@
                          "is_speaking": True,
                          "hotwords": hotword_msg,
                          "itn": use_itn,
                          })
        }
    )
    #voices.put(message)
    await websocket.send(message)
    while True:
@@ -159,6 +125,7 @@
        #voices.put(message)
        await websocket.send(message)
        await asyncio.sleep(0.005)


async def record_from_scp(chunk_begin, chunk_size):
    global voices
@@ -210,6 +177,7 @@
                audio_bytes = f.read()
        elif wav_path.endswith(".wav"):
            import wave

            with wave.open(wav_path, "rb") as wav_file:
                params = wav_file.getparams()
                sample_rate = wav_file.getframerate()
@@ -225,7 +193,9 @@
        # print(stride)

        # send first time
        message = json.dumps({"mode": args.mode,
        message = json.dumps(
            {
                "mode": args.mode,
                              "chunk_size": args.chunk_size,
                              "chunk_interval": args.chunk_interval,
                              "encoder_chunk_look_back": args.encoder_chunk_look_back,
@@ -235,7 +205,9 @@
                              "wav_format": wav_format,
                              "is_speaking": True,
                              "hotwords": hotword_msg,
                              "itn": use_itn})
                "itn": use_itn,
            }
        )

        #voices.put(message)
        await websocket.send(message)
@@ -253,7 +225,11 @@
                #voices.put(message)
                await websocket.send(message)
 
            sleep_duration = 0.001 if args.mode == "offline" else 60 * args.chunk_size[1] / args.chunk_interval / 1000
            sleep_duration = (
                0.001
                if args.mode == "offline"
                else 60 * args.chunk_size[1] / args.chunk_interval / 1000
            )
            
            await asyncio.sleep(sleep_duration)
    
@@ -269,14 +245,15 @@
    await websocket.close()


          
async def message(id):
    global websocket,voices,offline_msg_done
    text_print = ""
    text_print_2pass_online = ""
    text_print_2pass_offline = ""
    if args.output_dir is not None:
        ibest_writer = open(os.path.join(args.output_dir, "text.{}".format(id)), "a", encoding="utf-8")
        ibest_writer = open(
            os.path.join(args.output_dir, "text.{}".format(id)), "a", encoding="utf-8"
        )
    else:
        ibest_writer = None
    try:
@@ -298,12 +275,12 @@
                    text_write_line = "{}\t{}\n".format(wav_name, text)
                ibest_writer.write(text_write_line)

            if 'mode' not in meg:
            if "mode" not in meg:
                continue
            if meg["mode"] == "online":
                text_print += "{}".format(text)
                text_print = text_print[-args.words_max_print:]
                os.system('clear')
                os.system("clear")
                print("\rpid" + str(id) + ": " + text_print)
            elif meg["mode"] == "offline":
                if timestamp !="":
@@ -324,7 +301,7 @@
                    text_print = text_print_2pass_offline + "{}".format(text)
                    text_print_2pass_offline += "{}".format(text)
                text_print = text_print[-args.words_max_print:]
                os.system('clear')
                os.system("clear")
                print("\rpid" + str(id) + ": " + text_print)
                # offline_msg_done=True

@@ -332,8 +309,6 @@
            print("Exception:", e)
            #traceback.print_exc()
            #await websocket.close()
 



async def ws_client(id, chunk_begin, chunk_size):
@@ -354,7 +329,9 @@
        uri = "ws://{}:{}".format(args.host, args.port)
        ssl_context = None
    print("connect to", uri)
    async with websockets.connect(uri, subprotocols=["binary"], ping_interval=None, ssl=ssl_context) as websocket:
        async with websockets.connect(
            uri, subprotocols=["binary"], ping_interval=None, ssl=ssl_context
        ) as websocket:
        if args.audio_in is not None:
            task = asyncio.create_task(record_from_scp(i, 1))
        else:
@@ -368,13 +345,14 @@
    asyncio.get_event_loop().run_until_complete(ws_client(id, chunk_begin, chunk_size))
    asyncio.get_event_loop().run_forever()

if __name__ == '__main__':

if __name__ == "__main__":
    # for microphone
    if args.audio_in is None:
        p = Process(target=one_thread, args=(0, 0, 0))
        p.start()
        p.join()
        print('end')
        print("end")
    else:
        # calculate the number of wavs for each preocess
        if args.audio_in.endswith(".scp"):
@@ -387,7 +365,6 @@
            wav_name = wav_splits[0] if len(wav_splits) > 1 else "demo"
            wav_path = wav_splits[1] if len(wav_splits) > 1 else wav_splits[0]
            audio_type = os.path.splitext(wav_path)[-1].lower()


        total_len = len(wavs)
        if total_len >= args.thread_num:
@@ -413,4 +390,4 @@
        for i in process_list:
            p.join()

        print('end')
        print("end")